Research on complete coverage path planning for multiple unmanned underwater vehicles based on deep reinforcement learning

Swarm-style unmanned underwater vehicles (UUVs) possess significant application value in maritime search, early warning and reconnaissance, combat support, and defense operations. Complete coverage path planning (CCPP) represents a fundamental capability required for UUVs when executing missions such as search and detection. To address limitations in existing swarm-based CCPP research, including inadequate area partitioning methodologies and constrained path selection strategies, this paper establishes a comprehensive motion model for underwater vehicles. By treating autonomous underwater vehicles (AUVs) as intelligent agents, we introduce a novel multi-agent deep reinforcement learning framework. An enhanced A* algorithm is designed for pre-training, with the obtained solution serving as the initial input for deep reinforcement learning. The Hindsight Experience Replay method is incorporated to reconstruct the neural network’s training dataset, effectively resolving the sparse reward problem. Coordination and cooperation capabilities among agents are acquired through iterative learning. Simulation results demonstrate that compared to existing algorithms, our proposed approach significantly reduces coverage time and total path length while generating superior paths in unfamiliar environments. The algorithm demonstrates strong feasibility, adaptability, and practical applicability for real-world underwater missions.